BlueHive
BlueHive HUM
Agent setup
Remote MCPScoped accessStreamable HTTP

AI Agent Setup

Connect your AI assistant to HUM, then create agents, inspect calls, search phone inventory, manage campaigns, or provision organizations from your editor or terminal.

Browse agents

Prefer manual setup? Create a scoped API key

MCP endpoint
https://hum-api.bluehive.com/mcp
Authentication
Bearer API key
Tools
Agent lifecycle, search and execute
Start here

Connect in three steps

Create one least-privilege credential, connect your client, and verify discovery before asking an agent to change anything.

  1. 1

    Create a key

    In Settings → Integrations → API keys, choose the narrowest role and scopes that fit the work.

  2. 2

    Connect MCP

    Add the HUM endpoint and Bearer key to your agent using one of the configurations below.

  3. 3

    Verify safely

    Ask the agent to list operations or inspect current state before allowing a write action.

Choose a client

Pick your agent

Every option below supports HUM's remote MCP. Choose the workflow you already use.

Filter by workflow
Extension

GitHub Copilot

GitHub

Use HUM directly from Copilot Chat in agent mode with a securely prompted key.

Agent modeWorkspace contextSecure inputs
View setup
Terminal

Claude Code

Anthropic

Connect Claude Code through an environment-backed project MCP configuration.

Command lineProject configMCP tools
View setup
IDE

Cursor

Anysphere

Keep the key in your environment and share a safe project-level MCP definition.

Indexed codebaseEnvironment keysMCP tools
View setup
HTTP

Other MCP clients

Open standard

Connect any Streamable HTTP client that supports headers and Bearer credentials.

Streamable HTTPBearer authOpen standard
View setup
GitHub

GitHub Copilot

Use HUM directly from Copilot Chat in agent mode with a securely prompted key.

  1. 1Create .vscode/mcp.json in your project.
  2. 2Paste the configuration below. VS Code prompts for the key and stores it securely.
  3. 3Run MCP: List Servers, start bluehiveHum, then open Copilot Chat in agent mode.
.vscode/mcp.json
{
  "inputs": [
    {
      "type": "promptString",
      "id": "hum-api-key",
      "description": "BlueHive HUM API key",
      "password": true
    }
  ],
  "servers": {
    "bluehiveHum": {
      "type": "http",
      "url": "https://hum-api.bluehive.com/mcp",
      "headers": {
        "Authorization": "Bearer ${input:hum-api-key}"
      }
    }
  }
}
Anthropic

Claude Code

Connect Claude Code through an environment-backed project MCP configuration.

  1. 1Store your HUM key in the HUM_API_KEY environment variable or secret manager.
  2. 2Save the configuration below as .mcp.json in your project.
  3. 3Run claude mcp list to confirm bluehive-hum is connected.
.mcp.json
{
  "mcpServers": {
    "bluehive-hum": {
      "type": "http",
      "url": "https://hum-api.bluehive.com/mcp",
      "headers": {
        "Authorization": "Bearer ${HUM_API_KEY}"
      }
    }
  }
}
Anysphere

Cursor

Keep the key in your environment and share a safe project-level MCP definition.

  1. 1Export HUM_API_KEY in the shell that launches Cursor.
  2. 2Add the configuration below to .cursor/mcp.json.
  3. 3Open Customize → MCP and confirm the HUM tools are available.
.cursor/mcp.json
{
  "mcpServers": {
    "bluehive-hum": {
      "url": "https://hum-api.bluehive.com/mcp",
      "headers": {
        "Authorization": "Bearer ${env:HUM_API_KEY}"
      }
    }
  }
}
Open standard

Other MCP clients

Connect any Streamable HTTP client that supports headers and Bearer credentials.

  1. 1Choose Streamable HTTP as the transport.
  2. 2Set the HUM MCP URL and Authorization header shown below.
  3. 3If your client only supports stdio, bridge the remote endpoint with mcp-remote.
Connection details
URL: https://hum-api.bluehive.com/mcp
Transport: Streamable HTTP
Header: Authorization: Bearer hum_live_...
Accept: application/json, text/event-stream
Optional, recommended

Install the HUM skill

The skill teaches compatible agents HUM terminology, required safety checks, and reliable workflows without loading every API operation into context.

Terminal
npx skills add https://github.com/bluehive-health/hum-skills --skill bluehive-hum

The skill complements MCP: the skill explains how to work, while MCP provides the live tools and tenant-scoped data. Its source is published in the public HUM skills repository.

Compact by design

How HUM tools work

Named tools cover the agent lifecycle directly. Behind them, a generic pair reaches every remaining operation, generated from the current OpenAPI contract — so the whole API stays available without loading hundreds of definitions into context.

list_agents / get_agent

List the agents in your organization and inspect one, including its prompt, routing, and trigger events.

create_agent

Create a voice, SMS, or event agent in one step, without assembling a flow-configuration graph.

assign_agent_task

Hand an event agent a background task. Returns a run ID immediately rather than blocking on the work.

get_agent_run

Poll a run until it reports done, then read the agent's output, tool calls, and token usage.

search_hum_api

Find the exact operation ID, parameters, request schema, and confirmation requirement for a goal.

execute_hum_api

Execute one discovered operation through HUM's existing protected API handlers.

Good first prompt: List my HUM agents, then show me the schema needed to create a new inbound voice agent. Do not create anything yet.
Least privilege

Roles and scopes

A key combines an organization role with explicit scopes. Both must authorize an operation.

Use caseRoleScopes
Reporting and inspectionviewerorg:read
Normal operational workflowsoperatororg:read, org:write
Organization settingsadminorg:read, org:write
Cross-tenant provisioningadminorg:read, org:write, site:admin
Before production

Protect credentials and consequential actions

HUM keeps tenant, role, billing, consent, and validation checks authoritative even when an AI assistant initiates the request.

Never commit a HUM key

Store hum_live_* values in your client or secret manager. Revoke exposed keys immediately in Settings.

  • HUM requires explicit confirmation for purchases, outbound sends, launches, deletions, and provisioning.
  • API keys cannot create, enumerate, or revoke other API keys.
  • Treat transcripts, recordings, contacts, and patient context as sensitive data.
  • Respect plan limits, SMS consent, opt-outs, quiet hours, and tenant boundaries.